Merge tag 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost
[cris-mirror.git] / drivers / infiniband / hw / mthca / mthca_provider.c
blob6fee7795d1c8a8d58157c77099d3d80a96a19e7d
1 /*
2 * Copyright (c) 2004, 2005 Topspin Communications. All rights reserved.
3 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
4 * Copyright (c) 2005, 2006 Cisco Systems. All rights reserved.
5 * Copyright (c) 2005 Mellanox Technologies. All rights reserved.
6 * Copyright (c) 2004 Voltaire, Inc. All rights reserved.
8 * This software is available to you under a choice of one of two
9 * licenses. You may choose to be licensed under the terms of the GNU
10 * General Public License (GPL) Version 2, available from the file
11 * COPYING in the main directory of this source tree, or the
12 * OpenIB.org BSD license below:
14 * Redistribution and use in source and binary forms, with or
15 * without modification, are permitted provided that the following
16 * conditions are met:
18 * - Redistributions of source code must retain the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer.
22 * - Redistributions in binary form must reproduce the above
23 * copyright notice, this list of conditions and the following
24 * disclaimer in the documentation and/or other materials
25 * provided with the distribution.
27 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
28 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
29 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
30 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
31 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
32 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
33 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
34 * SOFTWARE.
37 #include <rdma/ib_smi.h>
38 #include <rdma/ib_umem.h>
39 #include <rdma/ib_user_verbs.h>
41 #include <linux/sched.h>
42 #include <linux/slab.h>
43 #include <linux/stat.h>
44 #include <linux/mm.h>
45 #include <linux/export.h>
47 #include "mthca_dev.h"
48 #include "mthca_cmd.h"
49 #include <rdma/mthca-abi.h>
50 #include "mthca_memfree.h"
52 static void init_query_mad(struct ib_smp *mad)
54 mad->base_version = 1;
55 mad->mgmt_class = IB_MGMT_CLASS_SUBN_LID_ROUTED;
56 mad->class_version = 1;
57 mad->method = IB_MGMT_METHOD_GET;
60 static int mthca_query_device(struct ib_device *ibdev, struct ib_device_attr *props,
61 struct ib_udata *uhw)
63 struct ib_smp *in_mad = NULL;
64 struct ib_smp *out_mad = NULL;
65 int err = -ENOMEM;
66 struct mthca_dev *mdev = to_mdev(ibdev);
68 if (uhw->inlen || uhw->outlen)
69 return -EINVAL;
71 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
72 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
73 if (!in_mad || !out_mad)
74 goto out;
76 memset(props, 0, sizeof *props);
78 props->fw_ver = mdev->fw_ver;
80 init_query_mad(in_mad);
81 in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
83 err = mthca_MAD_IFC(mdev, 1, 1,
84 1, NULL, NULL, in_mad, out_mad);
85 if (err)
86 goto out;
88 props->device_cap_flags = mdev->device_cap_flags;
89 props->vendor_id = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
90 0xffffff;
91 props->vendor_part_id = be16_to_cpup((__be16 *) (out_mad->data + 30));
92 props->hw_ver = be32_to_cpup((__be32 *) (out_mad->data + 32));
93 memcpy(&props->sys_image_guid, out_mad->data + 4, 8);
95 props->max_mr_size = ~0ull;
96 props->page_size_cap = mdev->limits.page_size_cap;
97 props->max_qp = mdev->limits.num_qps - mdev->limits.reserved_qps;
98 props->max_qp_wr = mdev->limits.max_wqes;
99 props->max_sge = mdev->limits.max_sg;
100 props->max_sge_rd = props->max_sge;
101 props->max_cq = mdev->limits.num_cqs - mdev->limits.reserved_cqs;
102 props->max_cqe = mdev->limits.max_cqes;
103 props->max_mr = mdev->limits.num_mpts - mdev->limits.reserved_mrws;
104 props->max_pd = mdev->limits.num_pds - mdev->limits.reserved_pds;
105 props->max_qp_rd_atom = 1 << mdev->qp_table.rdb_shift;
106 props->max_qp_init_rd_atom = mdev->limits.max_qp_init_rdma;
107 props->max_res_rd_atom = props->max_qp_rd_atom * props->max_qp;
108 props->max_srq = mdev->limits.num_srqs - mdev->limits.reserved_srqs;
109 props->max_srq_wr = mdev->limits.max_srq_wqes;
110 props->max_srq_sge = mdev->limits.max_srq_sge;
111 props->local_ca_ack_delay = mdev->limits.local_ca_ack_delay;
112 props->atomic_cap = mdev->limits.flags & DEV_LIM_FLAG_ATOMIC ?
113 IB_ATOMIC_HCA : IB_ATOMIC_NONE;
114 props->max_pkeys = mdev->limits.pkey_table_len;
115 props->max_mcast_grp = mdev->limits.num_mgms + mdev->limits.num_amgms;
116 props->max_mcast_qp_attach = MTHCA_QP_PER_MGM;
117 props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
118 props->max_mcast_grp;
120 * If Sinai memory key optimization is being used, then only
121 * the 8-bit key portion will change. For other HCAs, the
122 * unused index bits will also be used for FMR remapping.
124 if (mdev->mthca_flags & MTHCA_FLAG_SINAI_OPT)
125 props->max_map_per_fmr = 255;
126 else
127 props->max_map_per_fmr =
128 (1 << (32 - ilog2(mdev->limits.num_mpts))) - 1;
130 err = 0;
131 out:
132 kfree(in_mad);
133 kfree(out_mad);
134 return err;
137 static int mthca_query_port(struct ib_device *ibdev,
138 u8 port, struct ib_port_attr *props)
140 struct ib_smp *in_mad = NULL;
141 struct ib_smp *out_mad = NULL;
142 int err = -ENOMEM;
144 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
145 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
146 if (!in_mad || !out_mad)
147 goto out;
149 /* props being zeroed by the caller, avoid zeroing it here */
151 init_query_mad(in_mad);
152 in_mad->attr_id = IB_SMP_ATTR_PORT_INFO;
153 in_mad->attr_mod = cpu_to_be32(port);
155 err = mthca_MAD_IFC(to_mdev(ibdev), 1, 1,
156 port, NULL, NULL, in_mad, out_mad);
157 if (err)
158 goto out;
160 props->lid = be16_to_cpup((__be16 *) (out_mad->data + 16));
161 props->lmc = out_mad->data[34] & 0x7;
162 props->sm_lid = be16_to_cpup((__be16 *) (out_mad->data + 18));
163 props->sm_sl = out_mad->data[36] & 0xf;
164 props->state = out_mad->data[32] & 0xf;
165 props->phys_state = out_mad->data[33] >> 4;
166 props->port_cap_flags = be32_to_cpup((__be32 *) (out_mad->data + 20));
167 props->gid_tbl_len = to_mdev(ibdev)->limits.gid_table_len;
168 props->max_msg_sz = 0x80000000;
169 props->pkey_tbl_len = to_mdev(ibdev)->limits.pkey_table_len;
170 props->bad_pkey_cntr = be16_to_cpup((__be16 *) (out_mad->data + 46));
171 props->qkey_viol_cntr = be16_to_cpup((__be16 *) (out_mad->data + 48));
172 props->active_width = out_mad->data[31] & 0xf;
173 props->active_speed = out_mad->data[35] >> 4;
174 props->max_mtu = out_mad->data[41] & 0xf;
175 props->active_mtu = out_mad->data[36] >> 4;
176 props->subnet_timeout = out_mad->data[51] & 0x1f;
177 props->max_vl_num = out_mad->data[37] >> 4;
178 props->init_type_reply = out_mad->data[41] >> 4;
180 out:
181 kfree(in_mad);
182 kfree(out_mad);
183 return err;
186 static int mthca_modify_device(struct ib_device *ibdev,
187 int mask,
188 struct ib_device_modify *props)
190 if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
191 return -EOPNOTSUPP;
193 if (mask & IB_DEVICE_MODIFY_NODE_DESC) {
194 if (mutex_lock_interruptible(&to_mdev(ibdev)->cap_mask_mutex))
195 return -ERESTARTSYS;
196 memcpy(ibdev->node_desc, props->node_desc,
197 IB_DEVICE_NODE_DESC_MAX);
198 mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
201 return 0;
204 static int mthca_modify_port(struct ib_device *ibdev,
205 u8 port, int port_modify_mask,
206 struct ib_port_modify *props)
208 struct mthca_set_ib_param set_ib;
209 struct ib_port_attr attr;
210 int err;
212 if (mutex_lock_interruptible(&to_mdev(ibdev)->cap_mask_mutex))
213 return -ERESTARTSYS;
215 err = ib_query_port(ibdev, port, &attr);
216 if (err)
217 goto out;
219 set_ib.set_si_guid = 0;
220 set_ib.reset_qkey_viol = !!(port_modify_mask & IB_PORT_RESET_QKEY_CNTR);
222 set_ib.cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
223 ~props->clr_port_cap_mask;
225 err = mthca_SET_IB(to_mdev(ibdev), &set_ib, port);
226 if (err)
227 goto out;
228 out:
229 mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
230 return err;
233 static int mthca_query_pkey(struct ib_device *ibdev,
234 u8 port, u16 index, u16 *pkey)
236 struct ib_smp *in_mad = NULL;
237 struct ib_smp *out_mad = NULL;
238 int err = -ENOMEM;
240 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
241 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
242 if (!in_mad || !out_mad)
243 goto out;
245 init_query_mad(in_mad);
246 in_mad->attr_id = IB_SMP_ATTR_PKEY_TABLE;
247 in_mad->attr_mod = cpu_to_be32(index / 32);
249 err = mthca_MAD_IFC(to_mdev(ibdev), 1, 1,
250 port, NULL, NULL, in_mad, out_mad);
251 if (err)
252 goto out;
254 *pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
256 out:
257 kfree(in_mad);
258 kfree(out_mad);
259 return err;
262 static int mthca_query_gid(struct ib_device *ibdev, u8 port,
263 int index, union ib_gid *gid)
265 struct ib_smp *in_mad = NULL;
266 struct ib_smp *out_mad = NULL;
267 int err = -ENOMEM;
269 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
270 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
271 if (!in_mad || !out_mad)
272 goto out;
274 init_query_mad(in_mad);
275 in_mad->attr_id = IB_SMP_ATTR_PORT_INFO;
276 in_mad->attr_mod = cpu_to_be32(port);
278 err = mthca_MAD_IFC(to_mdev(ibdev), 1, 1,
279 port, NULL, NULL, in_mad, out_mad);
280 if (err)
281 goto out;
283 memcpy(gid->raw, out_mad->data + 8, 8);
285 init_query_mad(in_mad);
286 in_mad->attr_id = IB_SMP_ATTR_GUID_INFO;
287 in_mad->attr_mod = cpu_to_be32(index / 8);
289 err = mthca_MAD_IFC(to_mdev(ibdev), 1, 1,
290 port, NULL, NULL, in_mad, out_mad);
291 if (err)
292 goto out;
294 memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
296 out:
297 kfree(in_mad);
298 kfree(out_mad);
299 return err;
302 static struct ib_ucontext *mthca_alloc_ucontext(struct ib_device *ibdev,
303 struct ib_udata *udata)
305 struct mthca_alloc_ucontext_resp uresp;
306 struct mthca_ucontext *context;
307 int err;
309 if (!(to_mdev(ibdev)->active))
310 return ERR_PTR(-EAGAIN);
312 memset(&uresp, 0, sizeof uresp);
314 uresp.qp_tab_size = to_mdev(ibdev)->limits.num_qps;
315 if (mthca_is_memfree(to_mdev(ibdev)))
316 uresp.uarc_size = to_mdev(ibdev)->uar_table.uarc_size;
317 else
318 uresp.uarc_size = 0;
320 context = kmalloc(sizeof *context, GFP_KERNEL);
321 if (!context)
322 return ERR_PTR(-ENOMEM);
324 err = mthca_uar_alloc(to_mdev(ibdev), &context->uar);
325 if (err) {
326 kfree(context);
327 return ERR_PTR(err);
330 context->db_tab = mthca_init_user_db_tab(to_mdev(ibdev));
331 if (IS_ERR(context->db_tab)) {
332 err = PTR_ERR(context->db_tab);
333 mthca_uar_free(to_mdev(ibdev), &context->uar);
334 kfree(context);
335 return ERR_PTR(err);
338 if (ib_copy_to_udata(udata, &uresp, sizeof uresp)) {
339 mthca_cleanup_user_db_tab(to_mdev(ibdev), &context->uar, context->db_tab);
340 mthca_uar_free(to_mdev(ibdev), &context->uar);
341 kfree(context);
342 return ERR_PTR(-EFAULT);
345 context->reg_mr_warned = 0;
347 return &context->ibucontext;
350 static int mthca_dealloc_ucontext(struct ib_ucontext *context)
352 mthca_cleanup_user_db_tab(to_mdev(context->device), &to_mucontext(context)->uar,
353 to_mucontext(context)->db_tab);
354 mthca_uar_free(to_mdev(context->device), &to_mucontext(context)->uar);
355 kfree(to_mucontext(context));
357 return 0;
360 static int mthca_mmap_uar(struct ib_ucontext *context,
361 struct vm_area_struct *vma)
363 if (vma->vm_end - vma->vm_start != PAGE_SIZE)
364 return -EINVAL;
366 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
368 if (io_remap_pfn_range(vma, vma->vm_start,
369 to_mucontext(context)->uar.pfn,
370 PAGE_SIZE, vma->vm_page_prot))
371 return -EAGAIN;
373 return 0;
376 static struct ib_pd *mthca_alloc_pd(struct ib_device *ibdev,
377 struct ib_ucontext *context,
378 struct ib_udata *udata)
380 struct mthca_pd *pd;
381 int err;
383 pd = kmalloc(sizeof *pd, GFP_KERNEL);
384 if (!pd)
385 return ERR_PTR(-ENOMEM);
387 err = mthca_pd_alloc(to_mdev(ibdev), !context, pd);
388 if (err) {
389 kfree(pd);
390 return ERR_PTR(err);
393 if (context) {
394 if (ib_copy_to_udata(udata, &pd->pd_num, sizeof (__u32))) {
395 mthca_pd_free(to_mdev(ibdev), pd);
396 kfree(pd);
397 return ERR_PTR(-EFAULT);
401 return &pd->ibpd;
404 static int mthca_dealloc_pd(struct ib_pd *pd)
406 mthca_pd_free(to_mdev(pd->device), to_mpd(pd));
407 kfree(pd);
409 return 0;
412 static struct ib_ah *mthca_ah_create(struct ib_pd *pd,
413 struct rdma_ah_attr *ah_attr,
414 struct ib_udata *udata)
417 int err;
418 struct mthca_ah *ah;
420 ah = kmalloc(sizeof *ah, GFP_ATOMIC);
421 if (!ah)
422 return ERR_PTR(-ENOMEM);
424 err = mthca_create_ah(to_mdev(pd->device), to_mpd(pd), ah_attr, ah);
425 if (err) {
426 kfree(ah);
427 return ERR_PTR(err);
430 return &ah->ibah;
433 static int mthca_ah_destroy(struct ib_ah *ah)
435 mthca_destroy_ah(to_mdev(ah->device), to_mah(ah));
436 kfree(ah);
438 return 0;
441 static struct ib_srq *mthca_create_srq(struct ib_pd *pd,
442 struct ib_srq_init_attr *init_attr,
443 struct ib_udata *udata)
445 struct mthca_create_srq ucmd;
446 struct mthca_ucontext *context = NULL;
447 struct mthca_srq *srq;
448 int err;
450 if (init_attr->srq_type != IB_SRQT_BASIC)
451 return ERR_PTR(-ENOSYS);
453 srq = kmalloc(sizeof *srq, GFP_KERNEL);
454 if (!srq)
455 return ERR_PTR(-ENOMEM);
457 if (pd->uobject) {
458 context = to_mucontext(pd->uobject->context);
460 if (ib_copy_from_udata(&ucmd, udata, sizeof ucmd)) {
461 err = -EFAULT;
462 goto err_free;
465 err = mthca_map_user_db(to_mdev(pd->device), &context->uar,
466 context->db_tab, ucmd.db_index,
467 ucmd.db_page);
469 if (err)
470 goto err_free;
472 srq->mr.ibmr.lkey = ucmd.lkey;
473 srq->db_index = ucmd.db_index;
476 err = mthca_alloc_srq(to_mdev(pd->device), to_mpd(pd),
477 &init_attr->attr, srq);
479 if (err && pd->uobject)
480 mthca_unmap_user_db(to_mdev(pd->device), &context->uar,
481 context->db_tab, ucmd.db_index);
483 if (err)
484 goto err_free;
486 if (context && ib_copy_to_udata(udata, &srq->srqn, sizeof (__u32))) {
487 mthca_free_srq(to_mdev(pd->device), srq);
488 err = -EFAULT;
489 goto err_free;
492 return &srq->ibsrq;
494 err_free:
495 kfree(srq);
497 return ERR_PTR(err);
500 static int mthca_destroy_srq(struct ib_srq *srq)
502 struct mthca_ucontext *context;
504 if (srq->uobject) {
505 context = to_mucontext(srq->uobject->context);
507 mthca_unmap_user_db(to_mdev(srq->device), &context->uar,
508 context->db_tab, to_msrq(srq)->db_index);
511 mthca_free_srq(to_mdev(srq->device), to_msrq(srq));
512 kfree(srq);
514 return 0;
517 static struct ib_qp *mthca_create_qp(struct ib_pd *pd,
518 struct ib_qp_init_attr *init_attr,
519 struct ib_udata *udata)
521 struct mthca_create_qp ucmd;
522 struct mthca_qp *qp;
523 int err;
525 if (init_attr->create_flags)
526 return ERR_PTR(-EINVAL);
528 switch (init_attr->qp_type) {
529 case IB_QPT_RC:
530 case IB_QPT_UC:
531 case IB_QPT_UD:
533 struct mthca_ucontext *context;
535 qp = kmalloc(sizeof *qp, GFP_KERNEL);
536 if (!qp)
537 return ERR_PTR(-ENOMEM);
539 if (pd->uobject) {
540 context = to_mucontext(pd->uobject->context);
542 if (ib_copy_from_udata(&ucmd, udata, sizeof ucmd)) {
543 kfree(qp);
544 return ERR_PTR(-EFAULT);
547 err = mthca_map_user_db(to_mdev(pd->device), &context->uar,
548 context->db_tab,
549 ucmd.sq_db_index, ucmd.sq_db_page);
550 if (err) {
551 kfree(qp);
552 return ERR_PTR(err);
555 err = mthca_map_user_db(to_mdev(pd->device), &context->uar,
556 context->db_tab,
557 ucmd.rq_db_index, ucmd.rq_db_page);
558 if (err) {
559 mthca_unmap_user_db(to_mdev(pd->device),
560 &context->uar,
561 context->db_tab,
562 ucmd.sq_db_index);
563 kfree(qp);
564 return ERR_PTR(err);
567 qp->mr.ibmr.lkey = ucmd.lkey;
568 qp->sq.db_index = ucmd.sq_db_index;
569 qp->rq.db_index = ucmd.rq_db_index;
572 err = mthca_alloc_qp(to_mdev(pd->device), to_mpd(pd),
573 to_mcq(init_attr->send_cq),
574 to_mcq(init_attr->recv_cq),
575 init_attr->qp_type, init_attr->sq_sig_type,
576 &init_attr->cap, qp);
578 if (err && pd->uobject) {
579 context = to_mucontext(pd->uobject->context);
581 mthca_unmap_user_db(to_mdev(pd->device),
582 &context->uar,
583 context->db_tab,
584 ucmd.sq_db_index);
585 mthca_unmap_user_db(to_mdev(pd->device),
586 &context->uar,
587 context->db_tab,
588 ucmd.rq_db_index);
591 qp->ibqp.qp_num = qp->qpn;
592 break;
594 case IB_QPT_SMI:
595 case IB_QPT_GSI:
597 /* Don't allow userspace to create special QPs */
598 if (pd->uobject)
599 return ERR_PTR(-EINVAL);
601 qp = kmalloc(sizeof (struct mthca_sqp), GFP_KERNEL);
602 if (!qp)
603 return ERR_PTR(-ENOMEM);
605 qp->ibqp.qp_num = init_attr->qp_type == IB_QPT_SMI ? 0 : 1;
607 err = mthca_alloc_sqp(to_mdev(pd->device), to_mpd(pd),
608 to_mcq(init_attr->send_cq),
609 to_mcq(init_attr->recv_cq),
610 init_attr->sq_sig_type, &init_attr->cap,
611 qp->ibqp.qp_num, init_attr->port_num,
612 to_msqp(qp));
613 break;
615 default:
616 /* Don't support raw QPs */
617 return ERR_PTR(-ENOSYS);
620 if (err) {
621 kfree(qp);
622 return ERR_PTR(err);
625 init_attr->cap.max_send_wr = qp->sq.max;
626 init_attr->cap.max_recv_wr = qp->rq.max;
627 init_attr->cap.max_send_sge = qp->sq.max_gs;
628 init_attr->cap.max_recv_sge = qp->rq.max_gs;
629 init_attr->cap.max_inline_data = qp->max_inline_data;
631 return &qp->ibqp;
634 static int mthca_destroy_qp(struct ib_qp *qp)
636 if (qp->uobject) {
637 mthca_unmap_user_db(to_mdev(qp->device),
638 &to_mucontext(qp->uobject->context)->uar,
639 to_mucontext(qp->uobject->context)->db_tab,
640 to_mqp(qp)->sq.db_index);
641 mthca_unmap_user_db(to_mdev(qp->device),
642 &to_mucontext(qp->uobject->context)->uar,
643 to_mucontext(qp->uobject->context)->db_tab,
644 to_mqp(qp)->rq.db_index);
646 mthca_free_qp(to_mdev(qp->device), to_mqp(qp));
647 kfree(qp);
648 return 0;
651 static struct ib_cq *mthca_create_cq(struct ib_device *ibdev,
652 const struct ib_cq_init_attr *attr,
653 struct ib_ucontext *context,
654 struct ib_udata *udata)
656 int entries = attr->cqe;
657 struct mthca_create_cq ucmd;
658 struct mthca_cq *cq;
659 int nent;
660 int err;
662 if (attr->flags)
663 return ERR_PTR(-EINVAL);
665 if (entries < 1 || entries > to_mdev(ibdev)->limits.max_cqes)
666 return ERR_PTR(-EINVAL);
668 if (context) {
669 if (ib_copy_from_udata(&ucmd, udata, sizeof ucmd))
670 return ERR_PTR(-EFAULT);
672 err = mthca_map_user_db(to_mdev(ibdev), &to_mucontext(context)->uar,
673 to_mucontext(context)->db_tab,
674 ucmd.set_db_index, ucmd.set_db_page);
675 if (err)
676 return ERR_PTR(err);
678 err = mthca_map_user_db(to_mdev(ibdev), &to_mucontext(context)->uar,
679 to_mucontext(context)->db_tab,
680 ucmd.arm_db_index, ucmd.arm_db_page);
681 if (err)
682 goto err_unmap_set;
685 cq = kmalloc(sizeof *cq, GFP_KERNEL);
686 if (!cq) {
687 err = -ENOMEM;
688 goto err_unmap_arm;
691 if (context) {
692 cq->buf.mr.ibmr.lkey = ucmd.lkey;
693 cq->set_ci_db_index = ucmd.set_db_index;
694 cq->arm_db_index = ucmd.arm_db_index;
697 for (nent = 1; nent <= entries; nent <<= 1)
698 ; /* nothing */
700 err = mthca_init_cq(to_mdev(ibdev), nent,
701 context ? to_mucontext(context) : NULL,
702 context ? ucmd.pdn : to_mdev(ibdev)->driver_pd.pd_num,
703 cq);
704 if (err)
705 goto err_free;
707 if (context && ib_copy_to_udata(udata, &cq->cqn, sizeof (__u32))) {
708 mthca_free_cq(to_mdev(ibdev), cq);
709 err = -EFAULT;
710 goto err_free;
713 cq->resize_buf = NULL;
715 return &cq->ibcq;
717 err_free:
718 kfree(cq);
720 err_unmap_arm:
721 if (context)
722 mthca_unmap_user_db(to_mdev(ibdev), &to_mucontext(context)->uar,
723 to_mucontext(context)->db_tab, ucmd.arm_db_index);
725 err_unmap_set:
726 if (context)
727 mthca_unmap_user_db(to_mdev(ibdev), &to_mucontext(context)->uar,
728 to_mucontext(context)->db_tab, ucmd.set_db_index);
730 return ERR_PTR(err);
733 static int mthca_alloc_resize_buf(struct mthca_dev *dev, struct mthca_cq *cq,
734 int entries)
736 int ret;
738 spin_lock_irq(&cq->lock);
739 if (cq->resize_buf) {
740 ret = -EBUSY;
741 goto unlock;
744 cq->resize_buf = kmalloc(sizeof *cq->resize_buf, GFP_ATOMIC);
745 if (!cq->resize_buf) {
746 ret = -ENOMEM;
747 goto unlock;
750 cq->resize_buf->state = CQ_RESIZE_ALLOC;
752 ret = 0;
754 unlock:
755 spin_unlock_irq(&cq->lock);
757 if (ret)
758 return ret;
760 ret = mthca_alloc_cq_buf(dev, &cq->resize_buf->buf, entries);
761 if (ret) {
762 spin_lock_irq(&cq->lock);
763 kfree(cq->resize_buf);
764 cq->resize_buf = NULL;
765 spin_unlock_irq(&cq->lock);
766 return ret;
769 cq->resize_buf->cqe = entries - 1;
771 spin_lock_irq(&cq->lock);
772 cq->resize_buf->state = CQ_RESIZE_READY;
773 spin_unlock_irq(&cq->lock);
775 return 0;
778 static int mthca_resize_cq(struct ib_cq *ibcq, int entries, struct ib_udata *udata)
780 struct mthca_dev *dev = to_mdev(ibcq->device);
781 struct mthca_cq *cq = to_mcq(ibcq);
782 struct mthca_resize_cq ucmd;
783 u32 lkey;
784 int ret;
786 if (entries < 1 || entries > dev->limits.max_cqes)
787 return -EINVAL;
789 mutex_lock(&cq->mutex);
791 entries = roundup_pow_of_two(entries + 1);
792 if (entries == ibcq->cqe + 1) {
793 ret = 0;
794 goto out;
797 if (cq->is_kernel) {
798 ret = mthca_alloc_resize_buf(dev, cq, entries);
799 if (ret)
800 goto out;
801 lkey = cq->resize_buf->buf.mr.ibmr.lkey;
802 } else {
803 if (ib_copy_from_udata(&ucmd, udata, sizeof ucmd)) {
804 ret = -EFAULT;
805 goto out;
807 lkey = ucmd.lkey;
810 ret = mthca_RESIZE_CQ(dev, cq->cqn, lkey, ilog2(entries));
812 if (ret) {
813 if (cq->resize_buf) {
814 mthca_free_cq_buf(dev, &cq->resize_buf->buf,
815 cq->resize_buf->cqe);
816 kfree(cq->resize_buf);
817 spin_lock_irq(&cq->lock);
818 cq->resize_buf = NULL;
819 spin_unlock_irq(&cq->lock);
821 goto out;
824 if (cq->is_kernel) {
825 struct mthca_cq_buf tbuf;
826 int tcqe;
828 spin_lock_irq(&cq->lock);
829 if (cq->resize_buf->state == CQ_RESIZE_READY) {
830 mthca_cq_resize_copy_cqes(cq);
831 tbuf = cq->buf;
832 tcqe = cq->ibcq.cqe;
833 cq->buf = cq->resize_buf->buf;
834 cq->ibcq.cqe = cq->resize_buf->cqe;
835 } else {
836 tbuf = cq->resize_buf->buf;
837 tcqe = cq->resize_buf->cqe;
840 kfree(cq->resize_buf);
841 cq->resize_buf = NULL;
842 spin_unlock_irq(&cq->lock);
844 mthca_free_cq_buf(dev, &tbuf, tcqe);
845 } else
846 ibcq->cqe = entries - 1;
848 out:
849 mutex_unlock(&cq->mutex);
851 return ret;
854 static int mthca_destroy_cq(struct ib_cq *cq)
856 if (cq->uobject) {
857 mthca_unmap_user_db(to_mdev(cq->device),
858 &to_mucontext(cq->uobject->context)->uar,
859 to_mucontext(cq->uobject->context)->db_tab,
860 to_mcq(cq)->arm_db_index);
861 mthca_unmap_user_db(to_mdev(cq->device),
862 &to_mucontext(cq->uobject->context)->uar,
863 to_mucontext(cq->uobject->context)->db_tab,
864 to_mcq(cq)->set_ci_db_index);
866 mthca_free_cq(to_mdev(cq->device), to_mcq(cq));
867 kfree(cq);
869 return 0;
872 static inline u32 convert_access(int acc)
874 return (acc & IB_ACCESS_REMOTE_ATOMIC ? MTHCA_MPT_FLAG_ATOMIC : 0) |
875 (acc & IB_ACCESS_REMOTE_WRITE ? MTHCA_MPT_FLAG_REMOTE_WRITE : 0) |
876 (acc & IB_ACCESS_REMOTE_READ ? MTHCA_MPT_FLAG_REMOTE_READ : 0) |
877 (acc & IB_ACCESS_LOCAL_WRITE ? MTHCA_MPT_FLAG_LOCAL_WRITE : 0) |
878 MTHCA_MPT_FLAG_LOCAL_READ;
881 static struct ib_mr *mthca_get_dma_mr(struct ib_pd *pd, int acc)
883 struct mthca_mr *mr;
884 int err;
886 mr = kmalloc(sizeof *mr, GFP_KERNEL);
887 if (!mr)
888 return ERR_PTR(-ENOMEM);
890 err = mthca_mr_alloc_notrans(to_mdev(pd->device),
891 to_mpd(pd)->pd_num,
892 convert_access(acc), mr);
894 if (err) {
895 kfree(mr);
896 return ERR_PTR(err);
899 mr->umem = NULL;
901 return &mr->ibmr;
904 static struct ib_mr *mthca_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
905 u64 virt, int acc, struct ib_udata *udata)
907 struct mthca_dev *dev = to_mdev(pd->device);
908 struct scatterlist *sg;
909 struct mthca_mr *mr;
910 struct mthca_reg_mr ucmd;
911 u64 *pages;
912 int shift, n, len;
913 int i, k, entry;
914 int err = 0;
915 int write_mtt_size;
917 if (udata->inlen < sizeof ucmd) {
918 if (!to_mucontext(pd->uobject->context)->reg_mr_warned) {
919 mthca_warn(dev, "Process '%s' did not pass in MR attrs.\n",
920 current->comm);
921 mthca_warn(dev, " Update libmthca to fix this.\n");
923 ++to_mucontext(pd->uobject->context)->reg_mr_warned;
924 ucmd.mr_attrs = 0;
925 } else if (ib_copy_from_udata(&ucmd, udata, sizeof ucmd))
926 return ERR_PTR(-EFAULT);
928 mr = kmalloc(sizeof *mr, GFP_KERNEL);
929 if (!mr)
930 return ERR_PTR(-ENOMEM);
932 mr->umem = ib_umem_get(pd->uobject->context, start, length, acc,
933 ucmd.mr_attrs & MTHCA_MR_DMASYNC);
935 if (IS_ERR(mr->umem)) {
936 err = PTR_ERR(mr->umem);
937 goto err;
940 shift = mr->umem->page_shift;
941 n = mr->umem->nmap;
943 mr->mtt = mthca_alloc_mtt(dev, n);
944 if (IS_ERR(mr->mtt)) {
945 err = PTR_ERR(mr->mtt);
946 goto err_umem;
949 pages = (u64 *) __get_free_page(GFP_KERNEL);
950 if (!pages) {
951 err = -ENOMEM;
952 goto err_mtt;
955 i = n = 0;
957 write_mtt_size = min(mthca_write_mtt_size(dev), (int) (PAGE_SIZE / sizeof *pages));
959 for_each_sg(mr->umem->sg_head.sgl, sg, mr->umem->nmap, entry) {
960 len = sg_dma_len(sg) >> shift;
961 for (k = 0; k < len; ++k) {
962 pages[i++] = sg_dma_address(sg) + (k << shift);
964 * Be friendly to write_mtt and pass it chunks
965 * of appropriate size.
967 if (i == write_mtt_size) {
968 err = mthca_write_mtt(dev, mr->mtt, n, pages, i);
969 if (err)
970 goto mtt_done;
971 n += i;
972 i = 0;
977 if (i)
978 err = mthca_write_mtt(dev, mr->mtt, n, pages, i);
979 mtt_done:
980 free_page((unsigned long) pages);
981 if (err)
982 goto err_mtt;
984 err = mthca_mr_alloc(dev, to_mpd(pd)->pd_num, shift, virt, length,
985 convert_access(acc), mr);
987 if (err)
988 goto err_mtt;
990 return &mr->ibmr;
992 err_mtt:
993 mthca_free_mtt(dev, mr->mtt);
995 err_umem:
996 ib_umem_release(mr->umem);
998 err:
999 kfree(mr);
1000 return ERR_PTR(err);
1003 static int mthca_dereg_mr(struct ib_mr *mr)
1005 struct mthca_mr *mmr = to_mmr(mr);
1007 mthca_free_mr(to_mdev(mr->device), mmr);
1008 if (mmr->umem)
1009 ib_umem_release(mmr->umem);
1010 kfree(mmr);
1012 return 0;
1015 static struct ib_fmr *mthca_alloc_fmr(struct ib_pd *pd, int mr_access_flags,
1016 struct ib_fmr_attr *fmr_attr)
1018 struct mthca_fmr *fmr;
1019 int err;
1021 fmr = kmalloc(sizeof *fmr, GFP_KERNEL);
1022 if (!fmr)
1023 return ERR_PTR(-ENOMEM);
1025 memcpy(&fmr->attr, fmr_attr, sizeof *fmr_attr);
1026 err = mthca_fmr_alloc(to_mdev(pd->device), to_mpd(pd)->pd_num,
1027 convert_access(mr_access_flags), fmr);
1029 if (err) {
1030 kfree(fmr);
1031 return ERR_PTR(err);
1034 return &fmr->ibmr;
1037 static int mthca_dealloc_fmr(struct ib_fmr *fmr)
1039 struct mthca_fmr *mfmr = to_mfmr(fmr);
1040 int err;
1042 err = mthca_free_fmr(to_mdev(fmr->device), mfmr);
1043 if (err)
1044 return err;
1046 kfree(mfmr);
1047 return 0;
1050 static int mthca_unmap_fmr(struct list_head *fmr_list)
1052 struct ib_fmr *fmr;
1053 int err;
1054 struct mthca_dev *mdev = NULL;
1056 list_for_each_entry(fmr, fmr_list, list) {
1057 if (mdev && to_mdev(fmr->device) != mdev)
1058 return -EINVAL;
1059 mdev = to_mdev(fmr->device);
1062 if (!mdev)
1063 return 0;
1065 if (mthca_is_memfree(mdev)) {
1066 list_for_each_entry(fmr, fmr_list, list)
1067 mthca_arbel_fmr_unmap(mdev, to_mfmr(fmr));
1069 wmb();
1070 } else
1071 list_for_each_entry(fmr, fmr_list, list)
1072 mthca_tavor_fmr_unmap(mdev, to_mfmr(fmr));
1074 err = mthca_SYNC_TPT(mdev);
1075 return err;
1078 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
1079 char *buf)
1081 struct mthca_dev *dev =
1082 container_of(device, struct mthca_dev, ib_dev.dev);
1083 return sprintf(buf, "%x\n", dev->rev_id);
1086 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
1087 char *buf)
1089 struct mthca_dev *dev =
1090 container_of(device, struct mthca_dev, ib_dev.dev);
1091 switch (dev->pdev->device) {
1092 case PCI_DEVICE_ID_MELLANOX_TAVOR:
1093 return sprintf(buf, "MT23108\n");
1094 case PCI_DEVICE_ID_MELLANOX_ARBEL_COMPAT:
1095 return sprintf(buf, "MT25208 (MT23108 compat mode)\n");
1096 case PCI_DEVICE_ID_MELLANOX_ARBEL:
1097 return sprintf(buf, "MT25208\n");
1098 case PCI_DEVICE_ID_MELLANOX_SINAI:
1099 case PCI_DEVICE_ID_MELLANOX_SINAI_OLD:
1100 return sprintf(buf, "MT25204\n");
1101 default:
1102 return sprintf(buf, "unknown\n");
1106 static ssize_t show_board(struct device *device, struct device_attribute *attr,
1107 char *buf)
1109 struct mthca_dev *dev =
1110 container_of(device, struct mthca_dev, ib_dev.dev);
1111 return sprintf(buf, "%.*s\n", MTHCA_BOARD_ID_LEN, dev->board_id);
1114 static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
1115 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca, NULL);
1116 static DEVICE_ATTR(board_id, S_IRUGO, show_board, NULL);
1118 static struct device_attribute *mthca_dev_attributes[] = {
1119 &dev_attr_hw_rev,
1120 &dev_attr_hca_type,
1121 &dev_attr_board_id
1124 static int mthca_init_node_data(struct mthca_dev *dev)
1126 struct ib_smp *in_mad = NULL;
1127 struct ib_smp *out_mad = NULL;
1128 int err = -ENOMEM;
1130 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
1131 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
1132 if (!in_mad || !out_mad)
1133 goto out;
1135 init_query_mad(in_mad);
1136 in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
1138 err = mthca_MAD_IFC(dev, 1, 1,
1139 1, NULL, NULL, in_mad, out_mad);
1140 if (err)
1141 goto out;
1143 memcpy(dev->ib_dev.node_desc, out_mad->data, IB_DEVICE_NODE_DESC_MAX);
1145 in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
1147 err = mthca_MAD_IFC(dev, 1, 1,
1148 1, NULL, NULL, in_mad, out_mad);
1149 if (err)
1150 goto out;
1152 if (mthca_is_memfree(dev))
1153 dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
1154 memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
1156 out:
1157 kfree(in_mad);
1158 kfree(out_mad);
1159 return err;
1162 static int mthca_port_immutable(struct ib_device *ibdev, u8 port_num,
1163 struct ib_port_immutable *immutable)
1165 struct ib_port_attr attr;
1166 int err;
1168 immutable->core_cap_flags = RDMA_CORE_PORT_IBA_IB;
1170 err = ib_query_port(ibdev, port_num, &attr);
1171 if (err)
1172 return err;
1174 immutable->pkey_tbl_len = attr.pkey_tbl_len;
1175 immutable->gid_tbl_len = attr.gid_tbl_len;
1176 immutable->max_mad_size = IB_MGMT_MAD_SIZE;
1178 return 0;
1181 static void get_dev_fw_str(struct ib_device *device, char *str)
1183 struct mthca_dev *dev =
1184 container_of(device, struct mthca_dev, ib_dev);
1185 snprintf(str, IB_FW_VERSION_NAME_MAX, "%d.%d.%d",
1186 (int) (dev->fw_ver >> 32),
1187 (int) (dev->fw_ver >> 16) & 0xffff,
1188 (int) dev->fw_ver & 0xffff);
1191 int mthca_register_device(struct mthca_dev *dev)
1193 int ret;
1194 int i;
1196 ret = mthca_init_node_data(dev);
1197 if (ret)
1198 return ret;
1200 strlcpy(dev->ib_dev.name, "mthca%d", IB_DEVICE_NAME_MAX);
1201 dev->ib_dev.owner = THIS_MODULE;
1203 dev->ib_dev.uverbs_abi_ver = MTHCA_UVERBS_ABI_VERSION;
1204 dev->ib_dev.uverbs_cmd_mask =
1205 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) |
1206 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) |
1207 (1ull << IB_USER_VERBS_CMD_QUERY_PORT) |
1208 (1ull << IB_USER_VERBS_CMD_ALLOC_PD) |
1209 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) |
1210 (1ull << IB_USER_VERBS_CMD_REG_MR) |
1211 (1ull << IB_USER_VERBS_CMD_DEREG_MR) |
1212 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
1213 (1ull << IB_USER_VERBS_CMD_CREATE_CQ) |
1214 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ) |
1215 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) |
1216 (1ull << IB_USER_VERBS_CMD_CREATE_QP) |
1217 (1ull << IB_USER_VERBS_CMD_QUERY_QP) |
1218 (1ull << IB_USER_VERBS_CMD_MODIFY_QP) |
1219 (1ull << IB_USER_VERBS_CMD_DESTROY_QP) |
1220 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST) |
1221 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST);
1222 dev->ib_dev.node_type = RDMA_NODE_IB_CA;
1223 dev->ib_dev.phys_port_cnt = dev->limits.num_ports;
1224 dev->ib_dev.num_comp_vectors = 1;
1225 dev->ib_dev.dev.parent = &dev->pdev->dev;
1226 dev->ib_dev.query_device = mthca_query_device;
1227 dev->ib_dev.query_port = mthca_query_port;
1228 dev->ib_dev.modify_device = mthca_modify_device;
1229 dev->ib_dev.modify_port = mthca_modify_port;
1230 dev->ib_dev.query_pkey = mthca_query_pkey;
1231 dev->ib_dev.query_gid = mthca_query_gid;
1232 dev->ib_dev.alloc_ucontext = mthca_alloc_ucontext;
1233 dev->ib_dev.dealloc_ucontext = mthca_dealloc_ucontext;
1234 dev->ib_dev.mmap = mthca_mmap_uar;
1235 dev->ib_dev.alloc_pd = mthca_alloc_pd;
1236 dev->ib_dev.dealloc_pd = mthca_dealloc_pd;
1237 dev->ib_dev.create_ah = mthca_ah_create;
1238 dev->ib_dev.query_ah = mthca_ah_query;
1239 dev->ib_dev.destroy_ah = mthca_ah_destroy;
1241 if (dev->mthca_flags & MTHCA_FLAG_SRQ) {
1242 dev->ib_dev.create_srq = mthca_create_srq;
1243 dev->ib_dev.modify_srq = mthca_modify_srq;
1244 dev->ib_dev.query_srq = mthca_query_srq;
1245 dev->ib_dev.destroy_srq = mthca_destroy_srq;
1246 dev->ib_dev.uverbs_cmd_mask |=
1247 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ) |
1248 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ) |
1249 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ) |
1250 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ);
1252 if (mthca_is_memfree(dev))
1253 dev->ib_dev.post_srq_recv = mthca_arbel_post_srq_recv;
1254 else
1255 dev->ib_dev.post_srq_recv = mthca_tavor_post_srq_recv;
1258 dev->ib_dev.create_qp = mthca_create_qp;
1259 dev->ib_dev.modify_qp = mthca_modify_qp;
1260 dev->ib_dev.query_qp = mthca_query_qp;
1261 dev->ib_dev.destroy_qp = mthca_destroy_qp;
1262 dev->ib_dev.create_cq = mthca_create_cq;
1263 dev->ib_dev.resize_cq = mthca_resize_cq;
1264 dev->ib_dev.destroy_cq = mthca_destroy_cq;
1265 dev->ib_dev.poll_cq = mthca_poll_cq;
1266 dev->ib_dev.get_dma_mr = mthca_get_dma_mr;
1267 dev->ib_dev.reg_user_mr = mthca_reg_user_mr;
1268 dev->ib_dev.dereg_mr = mthca_dereg_mr;
1269 dev->ib_dev.get_port_immutable = mthca_port_immutable;
1270 dev->ib_dev.get_dev_fw_str = get_dev_fw_str;
1272 if (dev->mthca_flags & MTHCA_FLAG_FMR) {
1273 dev->ib_dev.alloc_fmr = mthca_alloc_fmr;
1274 dev->ib_dev.unmap_fmr = mthca_unmap_fmr;
1275 dev->ib_dev.dealloc_fmr = mthca_dealloc_fmr;
1276 if (mthca_is_memfree(dev))
1277 dev->ib_dev.map_phys_fmr = mthca_arbel_map_phys_fmr;
1278 else
1279 dev->ib_dev.map_phys_fmr = mthca_tavor_map_phys_fmr;
1282 dev->ib_dev.attach_mcast = mthca_multicast_attach;
1283 dev->ib_dev.detach_mcast = mthca_multicast_detach;
1284 dev->ib_dev.process_mad = mthca_process_mad;
1286 if (mthca_is_memfree(dev)) {
1287 dev->ib_dev.req_notify_cq = mthca_arbel_arm_cq;
1288 dev->ib_dev.post_send = mthca_arbel_post_send;
1289 dev->ib_dev.post_recv = mthca_arbel_post_receive;
1290 } else {
1291 dev->ib_dev.req_notify_cq = mthca_tavor_arm_cq;
1292 dev->ib_dev.post_send = mthca_tavor_post_send;
1293 dev->ib_dev.post_recv = mthca_tavor_post_receive;
1296 mutex_init(&dev->cap_mask_mutex);
1298 ret = ib_register_device(&dev->ib_dev, NULL);
1299 if (ret)
1300 return ret;
1302 for (i = 0; i < ARRAY_SIZE(mthca_dev_attributes); ++i) {
1303 ret = device_create_file(&dev->ib_dev.dev,
1304 mthca_dev_attributes[i]);
1305 if (ret) {
1306 ib_unregister_device(&dev->ib_dev);
1307 return ret;
1311 mthca_start_catas_poll(dev);
1313 return 0;
1316 void mthca_unregister_device(struct mthca_dev *dev)
1318 mthca_stop_catas_poll(dev);
1319 ib_unregister_device(&dev->ib_dev);